AI lowered the floor

Jonathan Wilke launched outbid.lol on 19 August 2026 at 11:08pm. The launch reached 1.6 million views. The product was one sentence: rank is the bid. Paste a URL, pay, sit where that number puts you.

instabid.lol is a clone of that mechanic, pointed at Instagram. We built it in days. Multiple other clones appeared in the same window.

Our clone reached 41 database rows and 5 visible listings.

Building it was the easy part. That is the whole argument. When anyone can ship a clone in a weekend, building is no longer a moat. Distribution is. We are the evidence.

What the floor actually is

Pieter Levels argued that indie hackers may be the first type of developer to go extinct as AI lowers the cost of execution. The frame is not that good builders get worse. It is that AI raises the ceiling and lowers the floor, so the space fills with people who can vibe-code their own tools.

The ceiling is familiar. A competent person with a model ships faster and attempts work that used to require a team. That part is not the shock.

The floor is the shock. The floor is the minimum time, skill, and money required to put a working product on the public internet.

For most of the last decade that minimum was high enough to act as a filter. You had to write the code, hire someone who could, or spend a quarter assembling a stack. Plenty of people had ideas. Fewer finished. Finishing was, by default, a kind of proof. It meant you had survived the expensive part.

That filter is gone. Indie Hackers observed that a weekend and a decent prompt now produces what used to take a quarter. Everyone is building. Everyone is building in public. The rooms where builders gather are louder than they have ever been.

Concretely, the floor is now a screenshot of a one-sentence product, a language model, a domain, and a payments provider if you need one. That is enough to clone a viral mechanic before the original has finished its first week.

We know because we did it. We did not invent ranking. We did not invent bidding. We pointed an existing sentence at a different object — a public Instagram handle instead of a URL — and had a board in days.

A lowered floor does two things at once. More people finish. Finishing stops being remarkable. If finishing is unremarkable, a shipped product is not evidence of a business. It is evidence that someone had a weekend.

The ceiling still exists. Taste, a harder problem, a better sentence, still produce a better thing. Levels is not saying the top of the distribution collapsed. He is saying the bottom rose until it is crowded. A crowd is not a moat. A crowd is noise.

The clone window

outbid.lol launched as a public object. A public object with 1.6 million views is a specification. Anyone who can read a screenshot can write the spec down.

Rank is the bid. The unit is whatever you paste in. The prize is being seen at the top of a page other people are already looking at.

That is not a deep product. It is a legible one. Legibility is what made it travel, and it is what made it cloneable. A product you can explain in one sentence is a product you can reimplement in one sitting.

We are not the only people who noticed. Multiple other clones appeared in the same window. Some changed the object. Some changed the stylesheet. Some copied the rules and hoped traffic would follow the domain.

instabid.lol is the Instagram version. The unit is a public handle. The ranking is still a bid. The page is still a list. The difference is the dataset, not the architecture.

If a clone of a mechanic can be built in days, the original mechanic was never the moat. The original’s traffic was. The first screenshot was. The audience of founders who were already the buyers was.

We cloned the mechanic. We did not clone 1.6 million views. We got 41 rows.

In an older market, a visible hit bought you months before a serious copy could exist. Here the copy arrived while the original was still the news. Concurrent copies do not compete on implementation quality in any way a visitor can see on first load. They compete on who is already being looked at.

Why a cloneable mechanic has no defensibility

A moat, in the old indie-hacker sense, was the difficulty of copying the work. If it took a quarter to ship, a copier was three months behind. Three months of traffic, customers, and taste was supposed to be enough.

A weekend collapses that gap to a rounding error. The copier is not behind. The copier is concurrent.

A cloneable mechanic has no defensibility for three structural reasons. None of them are about manners.

First, the rules are the product. Rank is the bid. There is no proprietary ranking function to hide, no trained model, no dataset a second board cannot start without. Anyone who has seen the page has the spec.

Second, the screenshot is the marketing, and the screenshot does not belong to the first person who took it. Once the format is in the feed — a numbered list, an amount, a handle — every clone can take the same picture. The format is not a brand. It is a genre.

Third, the switching cost is zero. A listing on one board does not lock a listing off another. A bidder who wants to be seen will bid where the eyes are. Eyes do not transfer when you copy the code.

The floor dropped. The room got louder. Nothing about that is a moat.

Defensibility would require something a weekend cannot copy: an audience that already trusts you, a channel you already own, data only you can see, a relationship with a buyer that is not “they saw a screenshot.” We had none of those. We had a mechanic and a domain.

The 41 rows are what a mechanic-plus-domain is worth on its own. Five of those rows became visible listings. The rest of the table is the distance between “it works” and “anyone wanted it.”

Even a perfect copy of the interface is not a copy of the graph underneath the original screenshot. The original board ranked products in front of founders. Founders were already looking, and founders were the buyers. We ranked Instagram handles in front of people looking at a clone of a founder product. The interface matched. The market did not.

Distribution is the remaining scarce asset

If building is cheap, the scarce input is the thing a better prompt does not make cheaper.

Distribution does not get cheaper. Attention is still allocated by platforms, by audiences that already exist, and by whether the people with the problem are in the room where you posted. A language model can write the landing page. It cannot put that page in front of a stranger who already wants the row.

In that same essay, Levels’s remainder is that the disruption can fertilize the soil for people who master distribution, branding, or proprietary data. That is not encouragement. It is an inventory. After execution is commoditized, those three are what is left.

We are the negative case. outbid.lol had distribution before it had a week of history. The launch went to an audience of founders. The founders were the customers. A late-night post could become a bid because the people watching were the people who buy rank on a page of startups.

We pointed the same mechanic at Instagram. Instagram’s users were not in that thread. The people who clone leaderboards are builders. The people who would fight for a row on an Instagram board are anyone whose handle is a public identity. Those groups do not overlap by default.

Indie Hackers put the distinction in the title: your build-in-public audience is not your market. The post describes a founder whose thread did 500 views and 50 comments in a day, then looked at who was in it: other founders, other builders, other indie hackers — not the freelancers the product was for. Applause measured the post. It did not measure demand.

Building was the easy part. It always is now.

1.6 million views belonged to the original launch. They were not a forecast of our listings. They were a count of how many people looked at a mechanic. Looking at a mechanic is how clones get made. It is not how Instagram accounts get ranked.

Distribution, used precisely, is not posting. Posting is how builders talk to builders. Distribution is being in the room where the buyer already has the problem. For outbid, that room was the indie timeline, and the problem was getting in front of other founders today. For instabid, that room would have been Instagram, or the places where handle vanity is already the sport. We did not start there. We started where the clone news was.

A 2026 review described build in public as having produced real careers and also a lot of theater. Theater is the right word for a launch that plays to other builders. The careers in that sentence were the cases where the audience and the buyer sat in the same room, or where the builder treated the posts as an archive and went to find customers somewhere else.

The remaining scarce asset is not a better prompt. It is a channel you did not have to rent on the day you shipped, because you already lived in it.

What we would do differently

Knowing what the 41 rows mean, the sequence would change.

We would not treat a 1.6 million-view object as a brief to reimplement. We would treat it as a brief to ask who the buyer is, and whether we can reach that buyer in days. If the answer is no, the clone is a writing exercise. We would still be allowed to write it. We would not be allowed to confuse it with a launch.

We would pick the room first. Instagram handles as the unit only works if Instagram users, or people who sell to them, can find the board without already being in a clone thread. That means starting on Instagram, or in the comments and group chats where handle rank is already contested, not on the timeline that was busy cloning .lol domains.

We would not spend the first days on fidelity to the original mechanic. Fidelity is what made us interchangeable with every other clone in the window. The useful work would have been the part a weekend of prompting cannot copy: a reason to be on this board and not the others, a relationship with a specific set of accounts, a path that does not depend on quoting the original.

We would seed the board in the market, not in the clone feed. Five visible listings is what you get when the people who see the product are mostly people who might build one. Density on a leaderboard is not a UI problem. It is a who-is-in-the-room problem. Empty rows are honest. They are also the screenshot that fails to travel.

We would measure demand before we measured completeness. Completeness is a site that accepts a handle and writes a row. Demand is a stranger who wants that row badly enough to finish. We had completeness in days. We had 5 visible listings. Those two facts belong on different dashboards.

We would still have shipped. “We cloned a mechanic in days” is a true sentence about cost. It is not a true sentence about a business.

What is still genuinely hard

The floor moved. The hard parts did not move with it.

It is still hard to be the first screenshot. outbid.lol had that. Every clone, including us, explained itself in relation to a page that already existed. First is a distribution event. It is not a coding event. You cannot prompt your way into being first after the picture has already been taken.

It is still hard to be in the buyer’s room. Builders are easy to talk to. They reply. They clone. They write about cloning. The buyer for an Instagram ranking is not in that conversation unless you put the product where the buyer already is. Finding that room, staying in it, and being useful there does not compress to a weekend.

It is still hard to make a reason to stay. A cloneable mechanic can acquire a listing. It cannot, by itself, make the listing worth defending. Switching cost is zero. The board that keeps people is the board the eyes are on, or the board that became a habit, or the board that means something the copies do not. Meaning is slow. Copies are fast.

It is still hard to tell theater from demand. Loudness is not a signal. The signal is a row a stranger completed without being in your replies. We have five of those. That number is more informative than the fact that the site exists.

It is still hard to own something a prompt cannot emit: a list, a name, data only you collected, relationships that predate the launch. We did not have those for Instagram, and the table looks like it.

Building was the easy part. The floor is down there. The work that is still work is everything that does not happen in the repository.